Clean Local AI Models on Mac Without Breaking Your Setup

Your Mac storage is filling up because of local AI. You know it. You want to fix it. But every article you find either tells you to delete entire folders with a terminal command or gives you a step-by-step guide that still leaves you unsure whether you deleted the right thing.

Local AI cleanup is not the same as deleting large files. It requires a different approach.

Why local AI storage is different

When you delete a large video file, the worst that happens is you lose the video. When you delete the wrong model blob, you might silently corrupt a model that three different tools depend on. When you delete an AI cache, you might remove project context that took months to build up. When you delete shared storage, you might break something you did not know was shared.

The risk profile is different. The cleanup approach needs to be different too.

What you are actually dealing with

Local AI storage on a typical Mac developer machine involves several distinct categories:

• Model weights — the actual files that make the AI work. Large. Sometimes shared between tools. Not rebuildable without re-downloading.
• Rebuildable caches — indexes, completion caches, and other derived data that can be regenerated. Safer to remove, but may affect performance until rebuilt.
• Project memory and context — AI context files tied to specific projects. May contain genuinely useful information. Requires review, not bulk deletion.
• Orphaned files — leftover files from models you have already removed through the normal interface. These are generally safe to clean but are invisible to the tools themselves.
• Duplicates — the same model downloaded through multiple tools and stored in multiple locations.

Treating all of these as "large files to delete" is how cleanups go wrong.

The steps that actually work

1. See everything before touching anything

Before deleting a single file, get a complete inventory. What tools have stored what, where, and how large is each item. Without this picture, you are making cleanup decisions based on incomplete information.

2. Reveal file paths before acting

Know exactly what you are about to remove. In Finder terms: see the actual file at its actual location before you decide to delete it. This sounds obvious, but most cleanup happens at a level of abstraction where you never see the actual file.

3. Move to Trash, not permanent delete

For model files especially, moving to Trash first means you can restore them if something breaks. The space is technically reserved until you empty the Trash, but you have a recovery window. Use it.

4. Export a report first

Before any significant cleanup, generate a record of what was on your machine. If something breaks weeks later and you cannot immediately identify why, having that inventory is invaluable.

5. Handle duplicates specifically

Duplicates are the highest-value cleanup target because you can reclaim significant space without losing any capability. But identifying them requires comparing across tools — not just looking within one application.

What not to do

A few approaches that seem reasonable but cause problems:

• Do not delete entire tool directories at once. They often contain a mix of rebuildable caches and non-rebuildable files.
• Do not delete blob files by searching for large files in Finder. Blob filenames give you no indication of what they belong to.
• Do not clean up while AI tools are running. Active tools may be writing to the files you are trying to remove.
• Do not skip the review step because you are in a hurry. Cleanup you regret takes much longer to fix than cleanup done carefully.